danielmorozoff

Definitely. I meant more recent than the foundations of information theory

Fundamentally, thermodynamics and information theory share a slew of similar ideas. Because in a lot of ways information theory came out of a thermodynamic thought process applied to signals. Even more recent work in ml like Boltzmann machines are directly taken from this overlap. That being said stronger theoretical connections are well worth the effort as these fields do have some different ideas that if shown to be similar can yield discoveries and push our understanding forward. Some thoughts that come to mind are compression, model generalizability/ convergence.

I highly recommend this book (talks) for anyone interested in a computation approach to neuroscience. I found it remarkable how far seeing Von Neumann was in relating computation to the understanding of the brain. Here is the physical copy if anyone is interested.

https://yalebooks.yale.edu/book/9780300181111/computer-and-b...

If you're interested in transcript search. We do it as part of our video processing engine and we work with a some podcast creators to provide search api. Message me if you are interested.

We do the first option in what you bring up in your later comment

Filecoin vc deal was the best imho executed vc play I have seen in a while. for 50m with a discount they were able to sell approx 200m to coin buyers. This not only offset their risk but literally gave the x4 in raw cash infusion in 2 weeks, growing the valuation without any time and this was done all on vc reputation/ name recognition/ and hype building. Very well done...

This is the current approach for large sale image retrieval. By using some model to extract features and then performing distance calculations. This is usually done with hashing once speed and the size of the dataset become large.

Are these monoclonals or polys? Does anyone have a paper ref?

First thing that came to mind

It would seem to me that this is an oversimplification. Doing screen based video processing is not simple and is not exact especially where camera movements are independent of game dynamics as well as the difficulties inherent in doing fast frame classification- there has been some good work in this direction recently but far from perfect(yolo, ssd). The route deep mind took in partnering with the game manufacturers to gain inputs directly while receiving global information about game state seems the simplest for their goal of training rl agents. We considered doing a project around this but figured it was a massive iceberg.

Vidrovr | Fullstack Engineer (FSE) | Full Time, $100k-$120k + equity. New York

We are a small engineering team who dropped out of our PhDs. We build video search and retrieval systems, leveraging various tech CV, ASR, NLP. We work with state of the art DL and PGM in these areas, with a a strong focus on large scale data processing, which is required to work with these models.

The FSE will take part in building out various system in the stack. Examples are:

front end web systems implemented in and Java, Python network and deployment systems leveraging various technologies (AWS/Jenkins/TeamCity) - Java, Python video/audio transcoding pipelines - Java, C++ DB deployments (SQL and noSQL, graph and key-val stores) Assisting in developing data collection and pre-processing pipelines for our video processing engine Engineering large data systems for video processing The FSE will manage the technical development of the product and work closely with business development to ensure that the technical design and aspects of the system meet business needs. All of the computing systems and interfaces will be built and maintained on Amazon Web Services.

email: contact@vidrovr.com

angel-list: https://angel.co/vidrovr/jobs/264959-full-stack-engineer

Frontend Designer/Developer | New York, USA | ONSITE

About Vidrovr: Vidrovr can index, search, and recommend video content in a cost-effective, automatic, and accurate manner. It was founded by Joseph Ellis and Daniel Morozoff, two former PhD Students in the Digital Video and Multimedia Lab at Columbia University and is advised by Prof. Shih-Fu Chang, of Columbia Engineering. The team has published and patented foundational research in machine learning, computer vision, multimodal information processing, and multimedia. Vidrovr addresses three key market needs: 1. Domain and customer specific automatic metadata generation for videos, 2. Video Content Management solutions that enable automatic placement and recommendation of video clips for digital products, and 3. Automatically linking and sourcing visual social media content that is relevant to a particular video or online article before it is published. Vidrovr is currently looking to build a team to commercialize and enhance the core technologies that have already been developed. Vidrovr was recently named as one of the winners of the prestigious Publicis90 competition, which entails investment and mentorship from Publicis Groupe. Vidrovr was awarded a National Science Foundation Small Business Innovation Research Grant to develop a unified multimodal framework for video understanding leveraging context and existing metadata. Vidrovr is backed by Techstars, and was part of the 2017 New York City Winter Cohort. Check out the Vidrovr website for more information. Please forward a CV with relevant publications/projects, work experience, and academic experience to contact [at] vidrovr.com if you are interested, include “Front End Engineer/ Designer Job Opening” in the title.

Lead design of the Vidrovr video management system web platform, which includes but is not limited to:

Front-end development utilizing latest technologies: HTML5/CSS3, React, Utilize backend data to pass information to front end. Familiarity with back end python/scala/java stacks Django 1.10 Play Framework 2 Jetty 9 User centered UI/UX design and implementations Create wireframes and UI/UX flows Implement designs in prototype mockups Prioritize feature additions Link feedback from users of the system

Designing the user workflow and interface for the following features: Vidrovr video search and indexing webpage Vidrovr video tools suite, including the iframe-embeds. Collaborate with management on the direction of the company’s secondary products using experience learned from interacting with users. Publication suite, social media to video linking, word press plug-in, etc. Market research and user studies to evaluate trends in the industry and help shape the near-term product roadmap of Vidrovr.

Can someone speak to how Agens compares to Neo4j? Since from a quick glance they seem to be using some of Neo's features.

I'd be interested in seeing performance comparisons as it seems Agens is implemented in C.

Vidrovr Inc. | Research Engineer | NYC | Full Time | http://vidrovr.com

About Vidrovr: Vidrovr can index, search, and recommend video content in a cost-effective, automatic, and accurate manner. It was founded by two former PhD Students in the Digital Video and Multimedia Lab at Columbia University and is advised by Prof. Shih-Fu Chang, of Columbia Engineering. The team has published and patented foundational research in machine learning, computer vision, multimodal information processing, and multimedia. Vidrovr addresses three key market needs: 1. Domain and customer specific automatic metadata generation for videos, 2. Video Content Management solutions that enable automatic placement and recommendation of video clips for digital products, and 3. Automatically linking and sourcing visual social media content that is relevant to a particular video or online article before it is published. Vidrovr is currently looking to build a team to commercialize and enhance the core technologies that have already been developed. Vidrovr just completed Techstars NY and was named as one of the winners of the prestigious Publicis90 competition, which entails investment and mentorship from Publicis Groupe. Vidrovr was awarded a National Science Foundation Small Business Innovation Research Grant to develop a unified multimodal framework for video understanding leveraging context and existing metadata. Check out the Vidrovr website for more information.

Job Description: The ML/CV Research Engineer position at Vidrovr can be described as follows: - The RE will take part in building a unified framework for processing and combining multimodal sources of information for video understanding. - The RE will lead the development real-time video and data analysis pipelines to handle large amounts of data, which could be social media streams, online articles, and/or images and video content. - The RE will manage the technical development of the product and work closely with business development to ensure that the technical design and aspects of the system meet business needs. - All of the computing systems and interfaces will be built and maintained on Amazon Web Services.

Job Requirements: - M.S. in CS, EE, Stats or other quantitative field (Ph.D. highly preferred) - 5+ years experience working with and designing large and real time data analytics systems, and a documented history of code or research accomplishments. - Proven research or documentable system development building state-of-the-art multimedia algorithms (Computer Vision, Data Mining, Machine Learning, etc.) - Highly proficient in at least one of the following programming languages: Python, C++, or Java. (Experience in Python preferred) - Friendly, hard-working, collaborative, and able to multitask are musts in an early-stage start-up, like Vidrovr. - The candidate must be excited about developing and being influential in creating the corporate culture at an early stage start-up, and implementing engineering policies and best practices, company-wide.

We are hiring for more generalist positions as well.

Feel free to reach out at contact@vidrovr.com

Thanks for the link. Are there any public specifics comparing distributed tf to Caffe2's distributed implementation?

Would also love some information on performance. How does Caffe2 stack up against TF1.0+? When is MKL support planned for ?

Once you get your feet wet, the first year PHD course book for good theory is Pattern Recognition and Machine Learning by Bishop.

But it could be a bit too theoretical - it provides a foundational mathematical framework and got me thinking about problems in a better way.

here is the nejm paper in the article: http://www.nejm.org/doi/10.1056/NEJMoa1609677

Yea seeing the same thing

Paper from Nature med: http://www.nature.com/nm/journal/vaop/ncurrent/full/nm.4268....

Here is the actual paper- it's remarkable.

http://science.sciencemag.org/content/early/2017/01/25/scien...

This was initially attempted by G-Cluster in 2000, Crytek tried it as well, popularized by OnLive (who started working on it in 2003) and blew up in 2010.

The general term for it is cloud gaming https://en.wikipedia.org/wiki/Cloud_gaming

That said, UCLA Math dept is one of the best in the country. Some very talented people work there.

There was a bunch of work done on bacteriophages by the soviets in the pre antibiotic era and then later as part of the bio weapons programs.

https://en.m.wikipedia.org/wiki/Phage_therapy

Here is the link to the theoretical paper: http://journals.aps.org/prd/abstract/10.1103/PhysRevD.94.101...

The cyberpunk future we were warned about is really coming to fruition.

If anyone is interested there was a huge body of work undertaken in the Soviet Union by a now mostly forgotten geneticist: Nikolai Vavilov. He has quite a story of persecution by the Soviet govt, and a great book on the topic may be found here:

https://play.google.com/store/books/details?id=2mi7NAOxe-8C&...

https://en.wikipedia.org/wiki/Nikolai_Vavilov

Here's the science paper: http://science.sciencemag.org/content/354/6314/878 reply

Here's the science paper: http://science.sciencemag.org/content/354/6314/878

That was an awesome description. Thank you

Though I totally understand your point.

I would argue relativity was a tectonic shift in how we understand physics. These ideas existed before Einstein in mathematics/ geometry (Grassmann & Reimann). But they were just ideas and not a full-fledged theory.